1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #ifndef _INET_IPSEC_IMPL_H 27 #define _INET_IPSEC_IMPL_H 28 29 #pragma ident "%Z%%M% %I% %E% SMI" 30 31 #include <inet/ip.h> 32 #include <inet/ipdrop.h> 33 34 #ifdef __cplusplus 35 extern "C" { 36 #endif 37 38 #define IPSEC_CONF_SRC_ADDRESS 0 /* Source Address */ 39 #define IPSEC_CONF_SRC_PORT 1 /* Source Port */ 40 #define IPSEC_CONF_DST_ADDRESS 2 /* Dest Address */ 41 #define IPSEC_CONF_DST_PORT 3 /* Dest Port */ 42 #define IPSEC_CONF_SRC_MASK 4 /* Source Address Mask */ 43 #define IPSEC_CONF_DST_MASK 5 /* Destination Address Mask */ 44 #define IPSEC_CONF_ULP 6 /* Upper layer Port */ 45 #define IPSEC_CONF_IPSEC_PROT 7 /* AH or ESP or AH_ESP */ 46 #define IPSEC_CONF_IPSEC_AALGS 8 /* Auth Algorithms - MD5 etc. */ 47 #define IPSEC_CONF_IPSEC_EALGS 9 /* Encr Algorithms - DES etc. */ 48 #define IPSEC_CONF_IPSEC_EAALGS 10 /* Encr Algorithms - MD5 etc. */ 49 #define IPSEC_CONF_IPSEC_SA 11 /* Shared or unique SA */ 50 #define IPSEC_CONF_IPSEC_DIR 12 /* Direction of traffic */ 51 #define IPSEC_CONF_ICMP_TYPE 13 /* ICMP type */ 52 #define IPSEC_CONF_ICMP_CODE 14 /* ICMP code */ 53 #define IPSEC_CONF_NEGOTIATE 15 /* Negotiation */ 54 #define IPSEC_CONF_TUNNEL 16 /* Tunnel */ 55 56 /* Type of an entry */ 57 58 #define IPSEC_NTYPES 0x02 59 #define IPSEC_TYPE_OUTBOUND 0x00 60 #define IPSEC_TYPE_INBOUND 0x01 61 62 /* Policy */ 63 #define IPSEC_POLICY_APPLY 0x01 64 #define IPSEC_POLICY_DISCARD 0x02 65 #define IPSEC_POLICY_BYPASS 0x03 66 67 /* Shared or unique SA */ 68 #define IPSEC_SHARED_SA 0x01 69 #define IPSEC_UNIQUE_SA 0x02 70 71 /* IPsec protocols and combinations */ 72 #define IPSEC_AH_ONLY 0x01 73 #define IPSEC_ESP_ONLY 0x02 74 #define IPSEC_AH_ESP 0x03 75 76 /* 77 * Internally defined "any" algorithm. 78 * Move to PF_KEY v3 when that RFC is released. 79 */ 80 #define SADB_AALG_ANY 255 81 82 #ifdef _KERNEL 83 84 #include <inet/common.h> 85 #include <netinet/ip6.h> 86 #include <netinet/icmp6.h> 87 #include <net/pfkeyv2.h> 88 #include <inet/ip.h> 89 #include <inet/sadb.h> 90 #include <inet/ipsecah.h> 91 #include <inet/ipsecesp.h> 92 #include <sys/crypto/common.h> 93 #include <sys/crypto/api.h> 94 #include <sys/avl.h> 95 96 /* 97 * Maximum number of authentication algorithms (can be indexed by one byte 98 * per PF_KEY and the IKE IPsec DOI. 99 */ 100 #define MAX_AALGS 256 101 102 /* 103 * IPsec task queue constants. 104 */ 105 #define IPSEC_TASKQ_MIN 10 106 #define IPSEC_TASKQ_MAX 20 107 108 /* 109 * So we can access IPsec global variables that live in keysock.c. 110 */ 111 extern boolean_t keysock_extended_reg(netstack_t *); 112 extern uint32_t keysock_next_seq(netstack_t *); 113 114 /* 115 * Locking for ipsec policy rules: 116 * 117 * policy heads: system policy is static; per-conn polheads are dynamic, 118 * and refcounted (and inherited); use atomic refcounts and "don't let 119 * go with both hands". 120 * 121 * policy: refcounted; references from polhead, ipsec_out 122 * 123 * actions: refcounted; referenced from: action hash table, policy, ipsec_out 124 * selectors: refcounted; referenced from: selector hash table, policy. 125 */ 126 127 /* 128 * the following are inspired by, but not directly based on, 129 * some of the sys/queue.h type-safe pseudo-polymorphic macros 130 * found in BSD. 131 * 132 * XXX If we use these more generally, we'll have to make the names 133 * less generic (HASH_* will probably clobber other namespaces). 134 */ 135 136 #define HASH_LOCK(table, hash) \ 137 mutex_enter(&(table)[hash].hash_lock) 138 #define HASH_UNLOCK(table, hash) \ 139 mutex_exit(&(table)[hash].hash_lock) 140 141 #define HASH_LOCKED(table, hash) \ 142 MUTEX_HELD(&(table)[hash].hash_lock) 143 144 #define HASH_ITERATE(var, field, table, hash) \ 145 var = table[hash].hash_head; var != NULL; var = var->field.hash_next 146 147 #define HASH_NEXT(var, field) \ 148 (var)->field.hash_next 149 150 #define HASH_INSERT(var, field, table, hash) \ 151 { \ 152 ASSERT(HASH_LOCKED(table, hash)); \ 153 (var)->field.hash_next = (table)[hash].hash_head; \ 154 (var)->field.hash_pp = &(table)[hash].hash_head; \ 155 (table)[hash].hash_head = var; \ 156 if ((var)->field.hash_next != NULL) \ 157 (var)->field.hash_next->field.hash_pp = \ 158 &((var)->field.hash_next); \ 159 } 160 161 162 #define HASH_UNCHAIN(var, field, table, hash) \ 163 { \ 164 ASSERT(MUTEX_HELD(&(table)[hash].hash_lock)); \ 165 HASHLIST_UNCHAIN(var, field); \ 166 } 167 168 #define HASHLIST_INSERT(var, field, head) \ 169 { \ 170 (var)->field.hash_next = head; \ 171 (var)->field.hash_pp = &(head); \ 172 head = var; \ 173 if ((var)->field.hash_next != NULL) \ 174 (var)->field.hash_next->field.hash_pp = \ 175 &((var)->field.hash_next); \ 176 } 177 178 #define HASHLIST_UNCHAIN(var, field) \ 179 { \ 180 *var->field.hash_pp = var->field.hash_next; \ 181 if (var->field.hash_next) \ 182 var->field.hash_next->field.hash_pp = \ 183 var->field.hash_pp; \ 184 HASH_NULL(var, field); \ 185 } 186 187 188 #define HASH_NULL(var, field) \ 189 { \ 190 var->field.hash_next = NULL; \ 191 var->field.hash_pp = NULL; \ 192 } 193 194 #define HASH_LINK(fieldname, type) \ 195 struct { \ 196 type *hash_next; \ 197 type **hash_pp; \ 198 } fieldname 199 200 201 #define HASH_HEAD(tag) \ 202 struct { \ 203 struct tag *hash_head; \ 204 kmutex_t hash_lock; \ 205 } 206 207 208 typedef struct ipsec_policy_s ipsec_policy_t; 209 210 typedef HASH_HEAD(ipsec_policy_s) ipsec_policy_hash_t; 211 212 /* 213 * When adding new fields to ipsec_prot_t, make sure to update 214 * ipsec_in_to_out_action() as well as other code in spd.c 215 */ 216 217 typedef struct ipsec_prot 218 { 219 unsigned int 220 ipp_use_ah : 1, 221 ipp_use_esp : 1, 222 ipp_use_se : 1, 223 ipp_use_unique : 1, 224 ipp_use_espa : 1, 225 ipp_pad : 27; 226 uint8_t ipp_auth_alg; /* DOI number */ 227 uint8_t ipp_encr_alg; /* DOI number */ 228 uint8_t ipp_esp_auth_alg; /* DOI number */ 229 uint16_t ipp_ah_minbits; /* AH: min keylen */ 230 uint16_t ipp_ah_maxbits; /* AH: max keylen */ 231 uint16_t ipp_espe_minbits; /* ESP encr: min keylen */ 232 uint16_t ipp_espe_maxbits; /* ESP encr: max keylen */ 233 uint16_t ipp_espa_minbits; /* ESP auth: min keylen */ 234 uint16_t ipp_espa_maxbits; /* ESP auth: max keylen */ 235 uint32_t ipp_km_proto; /* key mgmt protocol */ 236 uint32_t ipp_km_cookie; /* key mgmt cookie */ 237 uint32_t ipp_replay_depth; /* replay window */ 238 /* XXX add lifetimes */ 239 } ipsec_prot_t; 240 241 #define IPSEC_MAX_KEYBITS (0xffff) 242 243 /* 244 * An individual policy action, possibly a member of a chain. 245 * 246 * Action chains may be shared between multiple policy rules. 247 * 248 * With one exception (IPSEC_POLICY_LOG), a chain consists of an 249 * ordered list of alternative ways to handle a packet. 250 * 251 * All actions are also "interned" into a hash table (to allow 252 * multiple rules with the same action chain to share one copy in 253 * memory). 254 */ 255 256 typedef struct ipsec_act 257 { 258 uint8_t ipa_type; 259 uint8_t ipa_log; 260 union 261 { 262 ipsec_prot_t ipau_apply; 263 uint8_t ipau_reject_type; 264 uint32_t ipau_resolve_id; /* magic cookie */ 265 uint8_t ipau_log_type; 266 } ipa_u; 267 #define ipa_apply ipa_u.ipau_apply 268 #define ipa_reject_type ipa_u.ipau_reject_type 269 #define ipa_log_type ipa_u.ipau_log_type 270 #define ipa_resolve_type ipa_u.ipau_resolve_type 271 } ipsec_act_t; 272 273 #define IPSEC_ACT_APPLY 0x01 /* match IPSEC_POLICY_APPLY */ 274 #define IPSEC_ACT_DISCARD 0x02 /* match IPSEC_POLICY_DISCARD */ 275 #define IPSEC_ACT_BYPASS 0x03 /* match IPSEC_POLICY_BYPASS */ 276 #define IPSEC_ACT_REJECT 0x04 277 #define IPSEC_ACT_CLEAR 0x05 278 279 typedef struct ipsec_action_s 280 { 281 HASH_LINK(ipa_hash, struct ipsec_action_s); 282 struct ipsec_action_s *ipa_next; /* next alternative */ 283 uint32_t ipa_refs; /* refcount */ 284 ipsec_act_t ipa_act; 285 /* 286 * The following bits are equivalent to an OR of bits included in the 287 * ipau_apply fields of this and subsequent actions in an 288 * action chain; this is an optimization for the sake of 289 * ipsec_out_process() in ip.c and a few other places. 290 */ 291 unsigned int 292 ipa_hval: 8, 293 ipa_allow_clear:1, /* rule allows cleartext? */ 294 ipa_want_ah:1, /* an action wants ah */ 295 ipa_want_esp:1, /* an action wants esp */ 296 ipa_want_se:1, /* an action wants se */ 297 ipa_want_unique:1, /* want unique sa's */ 298 ipa_pad:19; 299 uint32_t ipa_ovhd; /* per-packet encap ovhd */ 300 } ipsec_action_t; 301 302 #define IPACT_REFHOLD(ipa) { \ 303 atomic_add_32(&(ipa)->ipa_refs, 1); \ 304 ASSERT((ipa)->ipa_refs != 0); \ 305 } 306 #define IPACT_REFRELE(ipa) { \ 307 ASSERT((ipa)->ipa_refs != 0); \ 308 membar_exit(); \ 309 if (atomic_add_32_nv(&(ipa)->ipa_refs, -1) == 0) \ 310 ipsec_action_free(ipa); \ 311 (ipa) = 0; \ 312 } 313 314 /* 315 * For now, use a trivially sized hash table for actions. 316 * In the future we can add the structure canonicalization necessary 317 * to get the hash function to behave correctly.. 318 */ 319 #define IPSEC_ACTION_HASH_SIZE 1 320 321 /* 322 * Merged address structure, for cheezy address-family independent 323 * matches in policy code. 324 */ 325 326 typedef union ipsec_addr 327 { 328 in6_addr_t ipsad_v6; 329 in_addr_t ipsad_v4; 330 } ipsec_addr_t; 331 332 /* 333 * ipsec selector set, as used by the kernel policy structures. 334 * Note that that we specify "local" and "remote" 335 * rather than "source" and "destination", which allows the selectors 336 * for symmetric policy rules to be shared between inbound and 337 * outbound rules. 338 * 339 * "local" means "destination" on inbound, and "source" on outbound. 340 * "remote" means "source" on inbound, and "destination" on outbound. 341 * XXX if we add a fifth policy enforcement point for forwarded packets, 342 * what do we do? 343 * 344 * The ipsl_valid mask is not done as a bitfield; this is so we 345 * can use "ffs()" to find the "most interesting" valid tag. 346 * 347 * XXX should we have multiple types for space-conservation reasons? 348 * (v4 vs v6? prefix vs. range)? 349 */ 350 351 typedef struct ipsec_selkey 352 { 353 uint32_t ipsl_valid; /* bitmask of valid entries */ 354 #define IPSL_REMOTE_ADDR 0x00000001 355 #define IPSL_LOCAL_ADDR 0x00000002 356 #define IPSL_REMOTE_PORT 0x00000004 357 #define IPSL_LOCAL_PORT 0x00000008 358 #define IPSL_PROTOCOL 0x00000010 359 #define IPSL_ICMP_TYPE 0x00000020 360 #define IPSL_ICMP_CODE 0x00000040 361 #define IPSL_IPV6 0x00000080 362 #define IPSL_IPV4 0x00000100 363 364 #define IPSL_WILDCARD 0x0000007f 365 366 ipsec_addr_t ipsl_local; 367 ipsec_addr_t ipsl_remote; 368 uint16_t ipsl_lport; 369 uint16_t ipsl_rport; 370 /* 371 * ICMP type and code selectors. Both have an end value to 372 * specify ranges, or * and *_end are equal for a single 373 * value 374 */ 375 uint8_t ipsl_icmp_type; 376 uint8_t ipsl_icmp_type_end; 377 uint8_t ipsl_icmp_code; 378 uint8_t ipsl_icmp_code_end; 379 380 uint8_t ipsl_proto; /* ip payload type */ 381 uint8_t ipsl_local_pfxlen; /* #bits of prefix */ 382 uint8_t ipsl_remote_pfxlen; /* #bits of prefix */ 383 uint8_t ipsl_mbz; 384 385 /* Insert new elements above this line */ 386 uint32_t ipsl_pol_hval; 387 uint32_t ipsl_sel_hval; 388 } ipsec_selkey_t; 389 390 typedef struct ipsec_sel 391 { 392 HASH_LINK(ipsl_hash, struct ipsec_sel); 393 uint32_t ipsl_refs; /* # refs to this sel */ 394 ipsec_selkey_t ipsl_key; /* actual selector guts */ 395 } ipsec_sel_t; 396 397 /* 398 * One policy rule. This will be linked into a single hash chain bucket in 399 * the parent rule structure. If the selector is simple enough to 400 * allow hashing, it gets filed under ipsec_policy_root_t->ipr_hash. 401 * Otherwise it goes onto a linked list in ipsec_policy_root_t->ipr_nonhash[af] 402 * 403 * In addition, we file the rule into an avl tree keyed by the rule index. 404 * (Duplicate rules are permitted; the comparison function breaks ties). 405 */ 406 struct ipsec_policy_s 407 { 408 HASH_LINK(ipsp_hash, struct ipsec_policy_s); 409 avl_node_t ipsp_byid; 410 uint64_t ipsp_index; /* unique id */ 411 uint32_t ipsp_prio; /* rule priority */ 412 uint32_t ipsp_refs; 413 ipsec_sel_t *ipsp_sel; /* selector set (shared) */ 414 ipsec_action_t *ipsp_act; /* action (may be shared) */ 415 }; 416 417 #define IPPOL_REFHOLD(ipp) { \ 418 atomic_add_32(&(ipp)->ipsp_refs, 1); \ 419 ASSERT((ipp)->ipsp_refs != 0); \ 420 } 421 #define IPPOL_REFRELE(ipp, ns) { \ 422 ASSERT((ipp)->ipsp_refs != 0); \ 423 membar_exit(); \ 424 if (atomic_add_32_nv(&(ipp)->ipsp_refs, -1) == 0) \ 425 ipsec_policy_free(ipp, ns); \ 426 (ipp) = 0; \ 427 } 428 429 #define IPPOL_UNCHAIN(php, ip, ns) \ 430 HASHLIST_UNCHAIN((ip), ipsp_hash); \ 431 avl_remove(&(php)->iph_rulebyid, (ip)); \ 432 IPPOL_REFRELE(ip, ns); 433 434 /* 435 * Policy ruleset. One per (protocol * direction) for system policy. 436 */ 437 438 #define IPSEC_AF_V4 0 439 #define IPSEC_AF_V6 1 440 #define IPSEC_NAF 2 441 442 typedef struct ipsec_policy_root_s 443 { 444 ipsec_policy_t *ipr_nonhash[IPSEC_NAF]; 445 int ipr_nchains; 446 ipsec_policy_hash_t *ipr_hash; 447 } ipsec_policy_root_t; 448 449 /* 450 * Policy head. One for system policy; there may also be one present 451 * on ill_t's with interface-specific policy, as well as one present 452 * for sockets with per-socket policy allocated. 453 */ 454 455 typedef struct ipsec_policy_head_s 456 { 457 uint32_t iph_refs; 458 krwlock_t iph_lock; 459 uint64_t iph_gen; /* generation number */ 460 ipsec_policy_root_t iph_root[IPSEC_NTYPES]; 461 avl_tree_t iph_rulebyid; 462 } ipsec_policy_head_t; 463 464 #define IPPH_REFHOLD(iph) { \ 465 atomic_add_32(&(iph)->iph_refs, 1); \ 466 ASSERT((iph)->iph_refs != 0); \ 467 } 468 #define IPPH_REFRELE(iph, ns) { \ 469 ASSERT((iph)->iph_refs != 0); \ 470 membar_exit(); \ 471 if (atomic_add_32_nv(&(iph)->iph_refs, -1) == 0) \ 472 ipsec_polhead_free(iph, ns); \ 473 (iph) = 0; \ 474 } 475 476 /* 477 * IPsec fragment related structures 478 */ 479 480 typedef struct ipsec_fragcache_entry { 481 struct ipsec_fragcache_entry *itpfe_next; /* hash list chain */ 482 mblk_t *itpfe_fraglist; /* list of fragments */ 483 time_t itpfe_exp; /* time when entry is stale */ 484 int itpfe_depth; /* # of fragments in list */ 485 ipsec_addr_t itpfe_frag_src; 486 ipsec_addr_t itpfe_frag_dst; 487 #define itpfe_src itpfe_frag_src.ipsad_v4 488 #define itpfe_src6 itpfe_frag_src.ipsad_v6 489 #define itpfe_dst itpfe_frag_dst.ipsad_v4 490 #define itpfe_dst6 itpfe_frag_dst.ipsad_v6 491 uint32_t itpfe_id; /* IP datagram ID */ 492 uint8_t itpfe_proto; /* IP Protocol */ 493 uint8_t itpfe_last; /* Last packet */ 494 } ipsec_fragcache_entry_t; 495 496 typedef struct ipsec_fragcache { 497 kmutex_t itpf_lock; 498 struct ipsec_fragcache_entry **itpf_ptr; 499 struct ipsec_fragcache_entry *itpf_freelist; 500 time_t itpf_expire_hint; /* time when oldest entry is stale */ 501 } ipsec_fragcache_t; 502 503 /* 504 * Tunnel policies. We keep a minature of the transport-mode/global policy 505 * per each tunnel instance. 506 * 507 * People who need both an itp held down AND one of its polheads need to 508 * first lock the itp, THEN the polhead, otherwise deadlock WILL occur. 509 */ 510 typedef struct ipsec_tun_pol_s { 511 avl_node_t itp_node; 512 kmutex_t itp_lock; 513 uint64_t itp_next_policy_index; 514 ipsec_policy_head_t *itp_policy; 515 ipsec_policy_head_t *itp_inactive; 516 uint32_t itp_flags; 517 uint32_t itp_refcnt; 518 char itp_name[LIFNAMSIZ]; 519 ipsec_fragcache_t itp_fragcache; 520 } ipsec_tun_pol_t; 521 /* NOTE - Callers (tun code) synchronize their own instances for these flags. */ 522 #define ITPF_P_ACTIVE 0x1 /* Are we using IPsec right now? */ 523 #define ITPF_P_TUNNEL 0x2 /* Negotiate tunnel-mode */ 524 /* Optimization -> Do we have per-port security entries in this polhead? */ 525 #define ITPF_P_PER_PORT_SECURITY 0x4 526 #define ITPF_PFLAGS 0x7 527 #define ITPF_SHIFT 3 528 529 #define ITPF_I_ACTIVE 0x8 /* Is the inactive using IPsec right now? */ 530 #define ITPF_I_TUNNEL 0x10 /* Negotiate tunnel-mode (on inactive) */ 531 /* Optimization -> Do we have per-port security entries in this polhead? */ 532 #define ITPF_I_PER_PORT_SECURITY 0x20 533 #define ITPF_IFLAGS 0x38 534 535 /* NOTE: f cannot be an expression. */ 536 #define ITPF_CLONE(f) (f) = (((f) & ITPF_PFLAGS) | \ 537 (((f) & ITPF_PFLAGS) << ITPF_SHIFT)); 538 #define ITPF_SWAP(f) (f) = ((((f) & ITPF_PFLAGS) << ITPF_SHIFT) | \ 539 (((f) & ITPF_IFLAGS) >> ITPF_SHIFT)) 540 541 #define ITP_P_ISACTIVE(itp, iph) ((itp)->itp_flags & \ 542 (((itp)->itp_policy == (iph)) ? ITPF_P_ACTIVE : ITPF_I_ACTIVE)) 543 544 #define ITP_P_ISTUNNEL(itp, iph) ((itp)->itp_flags & \ 545 (((itp)->itp_policy == (iph)) ? ITPF_P_TUNNEL : ITPF_I_TUNNEL)) 546 547 #define ITP_P_ISPERPORT(itp, iph) ((itp)->itp_flags & \ 548 (((itp)->itp_policy == (iph)) ? ITPF_P_PER_PORT_SECURITY : \ 549 ITPF_I_PER_PORT_SECURITY)) 550 551 #define ITP_REFHOLD(itp) { \ 552 atomic_add_32(&((itp)->itp_refcnt), 1); \ 553 ASSERT((itp)->itp_refcnt != 0); \ 554 } 555 556 #define ITP_REFRELE(itp, ns) { \ 557 ASSERT((itp)->itp_refcnt != 0); \ 558 membar_exit(); \ 559 if (atomic_add_32_nv(&((itp)->itp_refcnt), -1) == 0) \ 560 itp_free(itp, ns); \ 561 } 562 563 /* 564 * Certificate identity. 565 */ 566 567 typedef struct ipsid_s 568 { 569 struct ipsid_s *ipsid_next; 570 struct ipsid_s **ipsid_ptpn; 571 uint32_t ipsid_refcnt; 572 int ipsid_type; /* id type */ 573 char *ipsid_cid; /* certificate id string */ 574 } ipsid_t; 575 576 /* 577 * ipsid_t reference hold/release macros, just like ipsa versions. 578 */ 579 580 #define IPSID_REFHOLD(ipsid) { \ 581 atomic_add_32(&(ipsid)->ipsid_refcnt, 1); \ 582 ASSERT((ipsid)->ipsid_refcnt != 0); \ 583 } 584 585 /* 586 * Decrement the reference count on the ID. Someone else will clean up 587 * after us later. 588 */ 589 590 #define IPSID_REFRELE(ipsid) { \ 591 membar_exit(); \ 592 atomic_add_32(&(ipsid)->ipsid_refcnt, -1); \ 593 } 594 595 struct ipsec_out_s; 596 597 /* 598 * Following are the estimates of what the maximum AH and ESP header size 599 * would be. This is used to tell the upper layer the right value of MSS 600 * it should use without consulting AH/ESP. If the size is something 601 * different from this, ULP will learn the right one through 602 * ICMP_FRAGMENTATION_NEEDED messages generated locally. 603 * 604 * AH : 12 bytes of constant header + 32 bytes of ICV checksum (SHA-512). 605 */ 606 #define IPSEC_MAX_AH_HDR_SIZE (44) 607 608 /* 609 * ESP : Is a bit more complex... 610 * 611 * A system of one inequality and one equation MUST be solved for proper ESP 612 * overhead. The inequality is: 613 * 614 * MTU - sizeof (IP header + options) >= 615 * sizeof (esph_t) + sizeof (IV or ctr) + data-size + 2 + ICV 616 * 617 * IV or counter is almost always the cipher's block size. The equation is: 618 * 619 * data-size % block-size = (block-size - 2) 620 * 621 * so we can put as much data into the datagram as possible. If we are 622 * pessimistic and include our largest overhead cipher (AES) and hash 623 * (SHA-512), and assume 1500-byte MTU minus IPv4 overhead of 20 bytes, we get: 624 * 625 * 1480 >= 8 + 16 + data-size + 2 + 32 626 * 1480 >= 58 + data-size 627 * 1422 >= data-size, 1422 % 16 = 14, so 58 is the overhead! 628 * 629 * But, let's re-run the numbers with the same algorithms, but with an IPv6 630 * header: 631 * 632 * 1460 >= 58 + data-size 633 * 1402 >= data-size, 1402 % 16 = 10, meaning shrink to 1390 to get 14, 634 * 635 * which means the overhead is now 70. 636 * 637 * Hmmm... IPv4 headers can never be anything other than multiples of 4-bytes, 638 * and IPv6 ones can never be anything other than multiples of 8-bytes. We've 639 * seen overheads of 58 and 70. 58 % 16 == 10, and 70 % 16 == 6. IPv4 could 640 * force us to have 62 ( % 16 == 14) or 66 ( % 16 == 2), or IPv6 could force us 641 * to have 78 ( % 16 = 14). Let's compute IPv6 + 8-bytes of options: 642 * 643 * 1452 >= 58 + data-size 644 * 1394 >= data-size, 1394 % 16 = 2, meaning shrink to 1390 to get 14, 645 * 646 * Aha! The "ESP overhead" shrinks to 62 (70 - 8). This is good. Let's try 647 * IPv4 + 8 bytes of IPv4 options: 648 * 649 * 1472 >= 58 + data-size 650 * 1414 >= data-size, 1414 % 16 = 6, meaning shrink to 1406, 651 * 652 * meaning 66 is the overhead. Let's try 12 bytes: 653 * 654 * 1468 >= 58 + data-size 655 * 1410 >= data-size, 1410 % 16 = 2, meaning also shrink to 1406, 656 * 657 * meaning 62 is the overhead. How about 16 bytes? 658 * 659 * 1464 >= 58 + data-size 660 * 1406 >= data-size, 1402 % 16 = 14, which is great! 661 * 662 * this means 58 is the overhead. If I wrap and add 20 bytes, it looks just 663 * like IPv6's 70 bytes. If I add 24, we go back to 66 bytes. 664 * 665 * So picking 70 is a sensible, conservative default. Optimal calculations 666 * will depend on knowing pre-ESP header length (called "divpoint" in the ESP 667 * code), which could be cached in the conn_t for connected endpoints, or 668 * which must be computed on every datagram otherwise. 669 */ 670 #define IPSEC_MAX_ESP_HDR_SIZE (70) 671 672 /* 673 * Alternate, when we know the crypto block size via the SA. Assume an ICV on 674 * the SA. Use: 675 * 676 * sizeof (esph_t) + 2 * (sizeof (IV/counter)) - 2 + sizeof (ICV). The "-2" 677 * discounts the overhead of the pad + padlen that gets swallowed up by the 678 * second (theoretically all-pad) cipher-block. If you use our examples of 679 * AES and SHA512, you get: 680 * 681 * 8 + 32 - 2 + 32 == 70. 682 * 683 * Which is our pre-computed maximum above. 684 */ 685 #include <inet/ipsecesp.h> 686 #define IPSEC_BASE_ESP_HDR_SIZE(sa) \ 687 (sizeof (esph_t) + ((sa)->ipsa_iv_len << 1) - 2 + (sa)->ipsa_mac_len) 688 689 /* 690 * Identity hash table. 691 * 692 * Identities are refcounted and "interned" into the hash table. 693 * Only references coming from other objects (SA's, latching state) 694 * are counted in ipsid_refcnt. 695 * 696 * Locking: IPSID_REFHOLD is safe only when (a) the object's hash bucket 697 * is locked, (b) we know that the refcount must be > 0. 698 * 699 * The ipsid_next and ipsid_ptpn fields are only to be referenced or 700 * modified when the bucket lock is held; in particular, we only 701 * delete objects while holding the bucket lock, and we only increase 702 * the refcount from 0 to 1 while the bucket lock is held. 703 */ 704 705 #define IPSID_HASHSIZE 64 706 707 typedef struct ipsif_s 708 { 709 ipsid_t *ipsif_head; 710 kmutex_t ipsif_lock; 711 } ipsif_t; 712 713 714 /* 715 * IPsec stack instances 716 */ 717 struct ipsec_stack { 718 netstack_t *ipsec_netstack; /* Common netstack */ 719 720 /* Packet dropper for IP IPsec processing failures */ 721 ipdropper_t ipsec_dropper; 722 723 /* From spd.c */ 724 /* 725 * Policy rule index generator. We assume this won't wrap in the 726 * lifetime of a system. If we make 2^20 policy changes per second, 727 * this will last 2^44 seconds, or roughly 500,000 years, so we don't 728 * have to worry about reusing policy index values. 729 */ 730 uint64_t ipsec_next_policy_index; 731 732 HASH_HEAD(ipsec_action_s) ipsec_action_hash[IPSEC_ACTION_HASH_SIZE]; 733 HASH_HEAD(ipsec_sel) *ipsec_sel_hash; 734 uint32_t ipsec_spd_hashsize; 735 736 ipsif_t ipsec_ipsid_buckets[IPSID_HASHSIZE]; 737 738 /* 739 * Active & Inactive system policy roots 740 */ 741 ipsec_policy_head_t ipsec_system_policy; 742 ipsec_policy_head_t ipsec_inactive_policy; 743 744 /* Packet dropper for generic SPD drops. */ 745 ipdropper_t ipsec_spd_dropper; 746 krwlock_t ipsec_itp_get_byaddr_rw_lock; 747 ipsec_tun_pol_t *(*ipsec_itp_get_byaddr) 748 (uint32_t *, uint32_t *, int, netstack_t *); 749 750 /* ipdrop.c */ 751 kstat_t *ipsec_ip_drop_kstat; 752 struct ip_dropstats *ipsec_ip_drop_types; 753 754 /* spd.c */ 755 /* 756 * Have a counter for every possible policy message in 757 * ipsec_policy_failure_msgs 758 */ 759 uint32_t ipsec_policy_failure_count[IPSEC_POLICY_MAX]; 760 /* Time since last ipsec policy failure that printed a message. */ 761 hrtime_t ipsec_policy_failure_last; 762 763 /* ip_spd.c */ 764 /* stats */ 765 kstat_t *ipsec_ksp; 766 struct ipsec_kstats_s *ipsec_kstats; 767 768 /* sadb.c */ 769 /* Packet dropper for generic SADB drops. */ 770 ipdropper_t ipsec_sadb_dropper; 771 772 /* spd.c */ 773 boolean_t ipsec_inbound_v4_policy_present; 774 boolean_t ipsec_outbound_v4_policy_present; 775 boolean_t ipsec_inbound_v6_policy_present; 776 boolean_t ipsec_outbound_v6_policy_present; 777 778 /* spd.c */ 779 /* 780 * Because policy needs to know what algorithms are supported, keep the 781 * lists of algorithms here. 782 */ 783 kmutex_t ipsec_alg_lock; 784 785 uint8_t ipsec_nalgs[IPSEC_NALGTYPES]; 786 ipsec_alginfo_t *ipsec_alglists[IPSEC_NALGTYPES][IPSEC_MAX_ALGS]; 787 788 uint8_t ipsec_sortlist[IPSEC_NALGTYPES][IPSEC_MAX_ALGS]; 789 790 int ipsec_algs_exec_mode[IPSEC_NALGTYPES]; 791 792 uint32_t ipsec_tun_spd_hashsize; 793 /* 794 * Tunnel policies - AVL tree indexed by tunnel name. 795 */ 796 krwlock_t ipsec_tunnel_policy_lock; 797 uint64_t ipsec_tunnel_policy_gen; 798 avl_tree_t ipsec_tunnel_policies; 799 800 /* ipsec_loader.c */ 801 kmutex_t ipsec_loader_lock; 802 int ipsec_loader_state; 803 int ipsec_loader_sig; 804 kt_did_t ipsec_loader_tid; 805 kcondvar_t ipsec_loader_sig_cv; /* For loader_sig conditions. */ 806 807 }; 808 typedef struct ipsec_stack ipsec_stack_t; 809 810 /* Handle the kstat_create in ip_drop_init() failing */ 811 #define DROPPER(_ipss, _dropper) \ 812 (((_ipss)->ipsec_ip_drop_types == NULL) ? NULL : \ 813 &((_ipss)->ipsec_ip_drop_types->_dropper)) 814 815 /* 816 * Loader states.. 817 */ 818 #define IPSEC_LOADER_WAIT 0 819 #define IPSEC_LOADER_FAILED -1 820 #define IPSEC_LOADER_SUCCEEDED 1 821 822 /* 823 * ipsec_loader entrypoints. 824 */ 825 extern void ipsec_loader_init(ipsec_stack_t *); 826 extern void ipsec_loader_start(ipsec_stack_t *); 827 extern void ipsec_loader_destroy(ipsec_stack_t *); 828 extern void ipsec_loader_loadnow(ipsec_stack_t *); 829 extern boolean_t ipsec_loader_wait(queue_t *q, ipsec_stack_t *); 830 extern boolean_t ipsec_loaded(ipsec_stack_t *); 831 extern boolean_t ipsec_failed(ipsec_stack_t *); 832 833 /* 834 * callback from ipsec_loader to ip 835 */ 836 extern void ip_ipsec_load_complete(ipsec_stack_t *); 837 838 /* 839 * ipsec policy entrypoints (spd.c) 840 */ 841 842 extern void ipsec_policy_g_destroy(void); 843 extern void ipsec_policy_g_init(void); 844 845 extern int ipsec_alloc_table(ipsec_policy_head_t *, int, int, boolean_t, 846 netstack_t *); 847 extern void ipsec_polhead_init(ipsec_policy_head_t *, int); 848 extern void ipsec_polhead_destroy(ipsec_policy_head_t *); 849 extern void ipsec_polhead_free_table(ipsec_policy_head_t *); 850 extern mblk_t *ipsec_check_global_policy(mblk_t *, conn_t *, ipha_t *, 851 ip6_t *, boolean_t, netstack_t *); 852 extern mblk_t *ipsec_check_inbound_policy(mblk_t *, conn_t *, ipha_t *, ip6_t *, 853 boolean_t); 854 855 extern boolean_t ipsec_in_to_out(mblk_t *, ipha_t *, ip6_t *); 856 extern void ipsec_log_policy_failure(int, char *, ipha_t *, ip6_t *, boolean_t, 857 netstack_t *); 858 extern boolean_t ipsec_inbound_accept_clear(mblk_t *, ipha_t *, ip6_t *); 859 extern int ipsec_conn_cache_policy(conn_t *, boolean_t); 860 extern mblk_t *ipsec_alloc_ipsec_out(netstack_t *); 861 extern mblk_t *ipsec_attach_ipsec_out(mblk_t **, conn_t *, ipsec_policy_t *, 862 uint8_t, netstack_t *); 863 extern mblk_t *ipsec_init_ipsec_out(mblk_t *, mblk_t **, conn_t *, 864 ipsec_policy_t *, uint8_t, netstack_t *); 865 struct ipsec_in_s; 866 extern ipsec_action_t *ipsec_in_to_out_action(struct ipsec_in_s *); 867 extern boolean_t ipsec_check_ipsecin_latch(struct ipsec_in_s *, mblk_t *, 868 struct ipsec_latch_s *, ipha_t *, ip6_t *, const char **, kstat_named_t **, 869 conn_t *); 870 extern void ipsec_latch_inbound(ipsec_latch_t *ipl, struct ipsec_in_s *ii); 871 872 extern void ipsec_policy_free(ipsec_policy_t *, netstack_t *); 873 extern void ipsec_action_free(ipsec_action_t *); 874 extern void ipsec_polhead_free(ipsec_policy_head_t *, netstack_t *); 875 extern ipsec_policy_head_t *ipsec_polhead_split(ipsec_policy_head_t *, 876 netstack_t *); 877 extern ipsec_policy_head_t *ipsec_polhead_create(void); 878 extern ipsec_policy_head_t *ipsec_system_policy(netstack_t *); 879 extern ipsec_policy_head_t *ipsec_inactive_policy(netstack_t *); 880 extern void ipsec_swap_policy(ipsec_policy_head_t *, ipsec_policy_head_t *, 881 netstack_t *); 882 extern void ipsec_swap_global_policy(netstack_t *); 883 884 extern int ipsec_clone_system_policy(netstack_t *); 885 extern ipsec_policy_t *ipsec_policy_create(ipsec_selkey_t *, 886 const ipsec_act_t *, int, int, uint64_t *, netstack_t *); 887 extern boolean_t ipsec_policy_delete(ipsec_policy_head_t *, 888 ipsec_selkey_t *, int, netstack_t *); 889 extern int ipsec_policy_delete_index(ipsec_policy_head_t *, uint64_t, 890 netstack_t *); 891 extern void ipsec_polhead_flush(ipsec_policy_head_t *, netstack_t *); 892 extern int ipsec_copy_polhead(ipsec_policy_head_t *, ipsec_policy_head_t *, 893 netstack_t *); 894 extern void ipsec_actvec_from_req(ipsec_req_t *, ipsec_act_t **, uint_t *, 895 netstack_t *); 896 extern void ipsec_actvec_free(ipsec_act_t *, uint_t); 897 extern int ipsec_req_from_head(ipsec_policy_head_t *, ipsec_req_t *, int); 898 extern mblk_t *ipsec_construct_inverse_acquire(sadb_msg_t *, sadb_ext_t **, 899 netstack_t *); 900 extern mblk_t *ip_wput_attach_policy(mblk_t *, ipha_t *, ip6_t *, ire_t *, 901 conn_t *, boolean_t, zoneid_t); 902 extern mblk_t *ip_wput_ire_parse_ipsec_out(mblk_t *, ipha_t *, ip6_t *, 903 ire_t *, conn_t *, boolean_t, zoneid_t); 904 extern ipsec_policy_t *ipsec_find_policy(int, conn_t *, 905 struct ipsec_out_s *, ipsec_selector_t *, netstack_t *); 906 extern ipsid_t *ipsid_lookup(int, char *, netstack_t *); 907 extern boolean_t ipsid_equal(ipsid_t *, ipsid_t *); 908 extern void ipsid_gc(netstack_t *); 909 extern void ipsec_latch_ids(ipsec_latch_t *, ipsid_t *, ipsid_t *); 910 911 extern void ipsec_config_flush(netstack_t *); 912 extern boolean_t ipsec_check_policy(ipsec_policy_head_t *, ipsec_policy_t *, 913 int); 914 extern void ipsec_enter_policy(ipsec_policy_head_t *, ipsec_policy_t *, int, 915 netstack_t *); 916 extern boolean_t ipsec_check_action(ipsec_act_t *, int *, netstack_t *); 917 918 extern mblk_t *ipsec_out_tag(mblk_t *, mblk_t *, netstack_t *); 919 extern mblk_t *ipsec_in_tag(mblk_t *, mblk_t *, netstack_t *); 920 extern mblk_t *ip_copymsg(mblk_t *mp); 921 922 extern void iplatch_free(ipsec_latch_t *, netstack_t *); 923 extern ipsec_latch_t *iplatch_create(void); 924 extern int ipsec_set_req(cred_t *, conn_t *, ipsec_req_t *); 925 926 extern void ipsec_insert_always(avl_tree_t *tree, void *new_node); 927 928 extern int32_t ipsec_act_ovhd(const ipsec_act_t *act); 929 930 931 extern boolean_t iph_ipvN(ipsec_policy_head_t *, boolean_t); 932 933 /* 934 * Tunnel-support SPD functions and variables. 935 */ 936 struct tun_s; /* Defined in inet/tun.h. */ 937 extern boolean_t ipsec_tun_inbound(mblk_t *, mblk_t **, ipsec_tun_pol_t *, 938 ipha_t *, ip6_t *, ipha_t *, ip6_t *, int, netstack_t *); 939 extern mblk_t *ipsec_tun_outbound(mblk_t *, struct tun_s *, ipha_t *, 940 ip6_t *, ipha_t *, ip6_t *, int, netstack_t *); 941 extern void itp_free(ipsec_tun_pol_t *, netstack_t *); 942 extern ipsec_tun_pol_t *create_tunnel_policy(char *, int *, uint64_t *, 943 netstack_t *); 944 </